Batch ModelBuilder replication (#3415)

Signed-off-by: Alain Denzler <adenzler@nvidia.com>

Author: adenzler-nvidia

CHANGELOG.md

@@ -10,6 +10,7 @@
 
 - Compile tiled camera render kernels with CUDA fast math by default for faster rendering; set `SensorTiledCamera.render_config.enable_fast_math = False` for bit-exact, IEEE-precise output.
 - Optimize raycast/raytrace queries by restructuring ray-shape intersection into local-space primitives and compile specialized depth/shadow variants that skip unused surface-normal work (mesh shadows also use any-hit queries).
+- Speed up `ModelBuilder.replicate()` for large world counts by merging all copies in one pass; it no longer calls `add_world()` or `add_builder()` per copy, so `ModelBuilder` subclass overrides of those methods are not invoked during replication.
 
 ### Deprecated
 
@@ -19,13 +20,17 @@
 
 - Fix MJCF, URDF, and USD imports rendering collision-only bodies as visuals when the asset authors visual geometry elsewhere. (#3291)
 - Fix `ViewerUSD` texture consumers observing partially written PNGs by publishing generated textures atomically (#3288)
+- Fix builder merging (`ModelBuilder.add_builder()`, `add_world()`, `replicate()`) offsetting negative reference sentinels in custom attribute values stored as NumPy or Warp integer scalars.
 - Fix `ModelBuilder.add_usd()` requiring the optional `mujoco` package when handling `MjcActuator` prims, including during default MJC equality conversion.
 - Report malformed MJCF free-joint and inertial inputs with deterministic validation errors, and ignore MJCF mesh geom `size` lengths consistently.
 - Fix Style3D solver divergence caused by isolated vertices.
 - Fix the `diffsim_bear` example crashing with its default CUDA configuration and diverging after a few training iterations.
+- Preserve muscles and rigid-body color groups when copying or replicating a `ModelBuilder`.
 - Fix `ModelBuilder.add_usd()` to honor `PhysicsScene.gravityDirection`, including stage-to-builder rotation and per-world imports.
 - Fix stale overlay layers remaining visible after switching examples in the OpenGL viewer.
+- Reject incompatible custom attribute and frequency definitions before composing `ModelBuilder` instances.
 - Fix `cloth_franka` example rendering particles at simulation scale (cm) instead of viewer scale (m)
+- Fix `ModelBuilder` merges to accept array-valued transform fields and plain-list particle color groups.
 
 ## [1.4.0] - 2026-07-16